Service Dynamic Reconfiguration Strategy Based On Service Migration And Fog Cooperation

With the dramatic increase in Io T devices such as sensors,cars,drones,and robots,transferring large amounts of data to the cloud not only places a heavy burden on communication bandwidth,but also leads to degraded quality of service for users.There is therefore an urgent need to move the location of data processing from the cloud to the edge of the network closer to Io T devices,and Io T fog computing can be a good answer to these challenges.In Io T fog computing,due to some hardware limitations of the fog nodes,sometimes the fog nodes can only host limited services from the central cloud,yet the types of services requested by Io T devices are often highly diverse and may change over time,which may result in requests from Io T devices that cannot be processed by the local fog nodes in such cases.To ensure the timeliness of request processing,low energy consumption and other issues,fog nodes collaborate to assign task requests to other fog nodes or central cloud.Currently,there are two main ways for fog nodes to collaborate on task processing: horizontal offloading of tasks and vertical migration services.However,in the actual task request processing,the performance overheads of both horizontal offloading of data-intensive tasks and vertical migration of services from the central cloud to local fog nodes for service reconfiguration are large.Therefore,under the requirements of real-time,low-latency,and low-bandwidth Io T scenarios,neither of the above two collaborative approaches can process task requests efficiently and timely,resulting in degraded service quality.To address this challenge,this thesis introduces a computational paradigm of horizontal service migration and constructs a new collaborative task processing strategy around computational resources to bring the processing location of data closer to the data generation location,thereby reducing the overall network energy consumption and latency.The details of the research are as follows.1)Based on realistic task request patterns and task collaborative processing,we analyze the characteristics of several task request patterns,summarize a variety of request cases,and develop corresponding migration and offloading strategies.(2)A new inter-fog collaboration strategy is proposed,which introduces service horizontal migration while considering the traditional vertical migration and horizontal unloading collaboration methods,and performs task unloading and service migration in parallel,reducing the data transmission delay and minimizing the overall network energy consumption and response delay.(3)Through the analysis of multiple request scenarios,the problem of selecting task execution locations is modeled as a multidimensional Markovian decision process,and the deep reinforcement learning algorithm is used to achieve fast decision making for this problem by reasonably setting the state space,action space,reward and punishment functions and other attributes of the algorithm.To evaluate the feasibility and efficiency of this strategy,this thesis conducts simulation experiments on the constructed task co-processing problem under the fog computing network.The experimental results show that our proposed strategy can effectively reduce the overall network delay as well as energy loss.